Why Public Health Teams Should Think About AI as an Intern

If you've ever opened ChatGPT, pasted in a document, typed "make this better," and felt let down by what came back, this post is for you.

The problem usually isn't the tool. It's the mental model. If you're trying to figure out how to use AI in public health without it backfiring, getting that mental model right is the difference between AI that earns its keep and AI that gathers dust after the first week.

Most of us reach for AI like it's a vending machine. Put in a question, get out an answer. And when the answer is generic, or wrong, or weirdly confident about something it has no business being confident about, we decide the whole thing is overhyped and go back to doing it ourselves.

I want to offer you a different way to think about how to use AI in public health. Don't treat it like an answer machine. Treat it like an intern.

It sounds small, but it changes everything, because you already know how to work with an intern. You know they're capable but green. You know they need context before they're useful. You know to check their work before it goes out the door. That instinct is exactly what AI needs from you, and most people never think to apply it.

What an intern actually is

Think about the last time you had a bright, eager intern on your team. Maybe a student finishing their MPH. Smart, fast, genuinely wants to do good work. Knows the textbook cold.

But they don't know your county. They don't know which commissioner hates acronyms, or that the grant report is really due the Thursday before the long weekend, or that "engagement" means something very specific where you work. They've never sat in your meetings. So they'll hand you a confident first draft that completely misses the point. Not because they're not capable, but because you never told them what they needed to know.

That's AI. Capable, eager, fast, and operating with zero context about your world unless you give it some.

Once you hold that picture in your head, almost everything about working with AI gets easier.

You wouldn't say "feed me" at a restaurant

I heard a great line in one of our Community of Practice sessions. Walking up to a diner counter and saying "feed me" will get you something. Probably not what you wanted.

You'd never brief an intern that way. You wouldn't drop a 40-page policy on their desk, say "make it better," and walk off. You'd tell them what you're trying to accomplish, who it's for, what good looks like, and what to steer clear of. Something like: "Review this workshop script for high schoolers. I want students to feel comfortable talking about their mental health. Keep it conversational, skip the clinical jargon, and get it under two pages."

None of that is a magic prompt. It's context. I've watched a sharp person get a worse answer than a total beginner, and it almost always comes down to the same thing: they wrote a cleverer question but handed over the same thin slice of context.

When we train teams, we give them a simple way to remember how to brief their intern. We call it MAP.

Mapping the workflow comes first, and prompting comes last on purpose. Most people start with the prompt because it feels like the active part, but the step that actually moves the needle is the middle one. Adding context is just onboarding your intern.

A good intern needs guidelines (and so does your AI)

Here's the thing about a brand-new intern. On day one you don't just point at a desk and walk away. You give them guidelines. A job description so they know what they're responsible for. House rules for how your office does things. The shelf of approved guidance documents they're allowed to pull from. And a very clear line about what they're not allowed to touch, like the file with names and dates of birth in it.

You do all of that up front because it's cheaper than cleaning up after them later. The intern who's been given good guidelines does good work. The one you threw in cold writes a confident memo citing a policy that was rescinded two years ago.

AI is exactly the same, and most people skip the guidelines entirely. They open a blank chat, type one line, and expect a finished product. Then they're surprised when it reads like it was written by someone who's never set foot in their office. It was.

This is the "add the context" step from MAP, and it's where most of the payoff in public health AI actually lives. Here's what it looks like in practice:

• Write the job description. Tell it who it is and what it's for. "You're a health communications writer for a county health department. Your audience is residents reading at an eighth-grade level. You always write in plain language."
• Hand it the approved binder. Don't make it work from memory, because its memory is the whole internet and a lot of that is wrong. Give it your documents, your guidance, your real source material, and tell it to answer only from those.
• Set the house rules. "No jargon." "Under 200 words." "Cite your source." Constraints don't limit a good intern. They free them up to stop guessing.
• Draw the confidentiality line. This is the big one in public health AI. A general-purpose chatbot is an intern who never signed a confidentiality agreement. You cannot hand it protected health information, and in local government your chats may be part of the public record. Know what your intern is cleared to see before you hand it anything.

And then, like with any intern, you review the work. You would never take an intern's first draft, slap your name on it, and send it to the state. You read it. You catch the thing they got wrong because they didn't know any better. You hand it back: "This is close, but you missed the reporting requirement. Try again." AI doesn't know what it doesn't know, so your judgment stays the quality control. That part never goes away, and honestly, that's the good news.

What a good intern frees you up to do

The point of an intern was never to do your job. It was to take the pieces that don't need you, so you have room for the pieces that do.

I've watched this happen with environmental health inspectors. They used to lose real time digging through dense regulations to track down a single citation. Once they had an assistant that could pull it up with the source attached, the time saved wasn't even the interesting part. It was what they did with the room it opened up. They started using it to draft follow-up emails and corrective action plans, and they spent more of their day actually coaching on site.

That's the whole bargain. You hand the intern the part of the job that doesn't need your judgment, and you keep the part that does. No intern, human or AI, is going to read the room on a home visit, or earn a family's trust, or make the call that took you fifteen years of experience to learn to make. That's your work. The intern is just there so you have more time for it.

The trouble with a temp

Here's where the analogy gets uncomfortable for public health. A general-purpose AI tool is a temp. A sharp one, but a temp. It shows up with no clearance, no memory of yesterday, and no idea what your department actually does. Every single time you use it, you have to re-explain who you are, paste in your guidance again, and re-draw the confidentiality line you drew last week. And you still can't hand it anything sensitive, because it never signed the paperwork.

Re-explaining yourself every single time gets old fast, and it's a big part of why so many public health teams try AI once and quietly give up on it.

This is the problem we built PH360 to solve. It's purpose-built AI for public health, which is really just another way of saying it's the intern you've already onboarded. It understands the work, so nobody's explaining what an LHD is from scratch. It's HIPAA-compliant and keeps your data inside an environment your department controls, so the confidentiality line is built into the system instead of something you redraw every morning. You set the standing guidelines once, point it at your own approved documents, and it answers from those with the citation attached so you can check the source. That's what those inspectors had. Not a temp they had to brief every time, but an assistant that already understood their work and showed its sources.

None of that takes your judgment out of the loop. It just gives your judgment more to work with. That's the difference a purpose-built public health AI makes: you onboard it once instead of starting over every time you log in.

Where to start: how to use AI in public health

Here's the practical answer to how to use AI in public health: you don't need to automate your whole job. Pick one pain point. The thing that drains you, the search you run over and over, the same notes you key into three different systems. Hand just that piece to your "intern." Brief it like you'd brief a new hire, review what comes back like you'd review a draft, and redirect it when it's off.

And if you work inside a health department, you're not doing this alone or without rules. Find out what tools you already have. Copilot, Power Automate, whatever's sitting in your Microsoft license. Read your AI policy; you may already have one (and if you don't, we have a free template to start from). It also helps to see where public health AI fits the bigger picture, like the CDC's data modernization work. Then prototype something small and safe, and bring your supervisor evidence instead of just an idea.

The departments getting real value out of AI aren't the ones with the fanciest tools. They're the ones who got good at handing off work, checking what came back, and keeping their own judgment in charge of the final call.

That's a skill your team already has. You've been managing interns for years.